Spire Global unveils its AI-S2S Model, with groundbreaking long-range weather forecasting

This state-of-the-art sub-seasonal to seasonal (S2S) forecasting system stands apart from other AI models in the market, thanks to its probabilistic forecasts generating 200 ensemble members, coupled with Spire’s exclusive satellite data assimilation. The AI-S2S model delivers unparalleled long-range weather insights, helping businesses anticipate weather-driven risks and opportunities with greater confidence.

Spire’s AI-S2S model is the result of years of expertise from a diverse team of scientists, blending deep learning, meteorology, and physics to push the boundaries of long-range forecasting.

Bridging the ‘valley of unpredictability’ with AI innovation in long-range weather forecasting

Sub-seasonal forecasting has traditionally been a formidable challenge, often referred to as the “valley of unpredictability” in meteorology. Forecast accuracy significantly declines beyond the 10-15-day range, leaving industries vulnerable to weather disruptions. Spire’s AI-S2S model disrupts this paradigm by leveraging deep learning techniques, probabilistic forecasting, and Spire’s exclusive satellite data to push the boundaries of predictability up to 45 days in advance.

“Our model is uniquely designed to capture long-range weather dependencies, integrating slow-evolving variables like sea surface temperature (SST) and soil moisture to enhance forecast reliability,” said Dr. Nachiketa Acharya, Senior AI Weather and Climate Scientist at Spire Global. “This level of accuracy and foresight has been largely unattainable — until now.”

A snapshot of 240-hour, 2-meter temperature forecasts is shown for 12 ensemble members generated by Spire Global’s AI-S2S model, which has 200 ensemble members operationally.

Dr. Oyebade Oyedotun, a Senior Machine Learning Scientist at Spire, holds a PhD in computer science from the University of Luxembourg, where he specialized in machine learning and computer vision. “Traditional numerical models rely on solving complex equations with many assumptions,” Dr. Oyedotun said. “AI, on the other hand, learns patterns directly from data, making it more flexible and adaptive for long-range forecasting.”

Why AI-S2S is revolutionary in long-range weather forecasting

Most existing AI weather models operate on deterministic frameworks, providing a single, best-guess forecast. Spire’s AI-S2S model, however, takes a fundamentally different approach by offering a probabilistic forecast with 200 ensemble members. This means businesses don’t just get one potential outcome — they receive a distribution of possibilities, allowing for better risk assessment and strategic planning, leading to better preparation for extreme weather events.

Dr. Oyedotun explains: “Weather is inherently chaotic, and small differences in initial conditions can lead to vastly different outcomes. By running 200 ensemble members, our model quantifies uncertainty with exceptional granularity, providing industries with a much clearer picture of potential scenarios.”

Powered by Nvidia GPUs, Spire’s AI models run 1,000 times faster than traditional physics-based models, enabling large ensemble forecasts that capture the full range of possible weather outcomes.

Dr. Acharya, who earned a PhD in statistics from Utkal University, India, has spent more than 15 years of his career refining sub-seasonal and seasonal forecasting methods, focused on real-world applications. “By embedding probabilistic approaches directly into our AI model instead of just running deterministic models with different initial conditions, we can quantify uncertainty in ways never before possible,” Acharya said.

Satellite data assimilation: Spire’s unique strength for enhancing forecast accuracy

A major differentiator of the AI-S2S model is its integration of Spire’s proprietary satellite data. Unlike conventional AI models that rely on publicly available reanalysis datasets, Spire’s satellite constellation provides real-time, high-resolution data on critical atmospheric and environmental variables.

“By training our AI model with exclusive satellite data — such as Radio Occultation (RO) measurements and GNSS-R derived soil moisture and ocean surface winds — we ensure that our forecasts are rooted in the most up-to-date and accurate information available,” said Dr. Luis Vela, Senior AI Weather Scientist at Spire. With a PhD spanning research across Germany, Australia, the United States, and Spain, Vela’s background in computational physics and numerical simulations helps optimize the infrastructure that powers Spire’s AI-S2S model. “This allows us to significantly improve forecast reliability beyond what traditional models can achieve.”

Daily mean inputs are also used instead of instantaneous values, enhancing forecast accuracy for long-range predictions.

“In sub-seasonal forecasting, we’re more interested in large-scale weather patterns over time rather than fine details. Using daily mean values rather than instantaneous six-hour intervals helps smooth out short-term fluctuations that aren’t relevant to long-term trends. By using daily means, we filter out short-term noise and focus on persistent weather patterns, improving long-range forecast stability,” Oyedotun said.

Accurate initial conditions are critical for high-quality forecasts, but for sub-seasonal predictions, they aren’t enough.

“You also need a strong modeling system that can accurately capture how these conditions evolve over time,” Dr. Oyedotun explained.

“For example, if you want to predict weather 20 days from now, your system needs to process the initial conditions and project them forward in a way that accurately reflects real-world atmospheric evolution,” he added. “That’s where AI comes in. We’re leveraging state-of-the-art AI techniques to train models that not only start with good initial conditions but also provide high-quality long-range forecasts.”

A snapshot of 240-hour, 100-meter wind speed forecasts is shown for 12 ensemble members generated by Spire Global’s AI-S2S model, which has 200 ensemble members operationally.

Key industry applications: Enabling better decisions for energy, agriculture, and finance

“The scale of this model’s impact is immense,” said Mike Eilts, General Manager of Spire Weather & Climate. “From predicting energy demand fluctuations to helping farmers anticipate drought risks, AI-S2S provides insights that drive smarter decision-making across industries that have long sought out actionable sub-seasonal forecasts. With Spire’s AI-S2S model, we’ve turned that vision into reality.”

Spire’s AI-S2S model is poised to revolutionize multiple industries by delivering useful, long-range weather forecasting insights:

• Energy trading and commodities — Traders can anticipate temperature-driven energy demand shifts, market fluctuations, and optimize pricing strategies based on probabilistic weather outcomes
• Agriculture and food security — Farmers can better plan for drought risks, precipitation trends, and temperature variations to inform planting, irrigation, pest and disease management, and harvesting decisions
• Financial markets and risk management — Investors and insurers can hedge financial risk tied to extreme weather, allowing for refined pricing of policies by assessing long-term risk
• Supply chain and logistics — Companies can proactively adjust logistics and inventory planning in response to predicted weather disruptions such as adjusting shipping routes when severe weather is forecast

“Beyond energy and agriculture, S2S forecasting has critical applications in disaster preparedness,” Dr. Vela added. “Predicting droughts, floods, and wildfires weeks in advance can help governments and businesses mitigate risks. Climate change is making extreme weather more frequent, so having a reliable long-range forecast can be a game changer for resilience planning.”

Spire’s commitment to AI leadership in weather forecasting

Spire is setting a new benchmark in AI-driven weather intelligence. Unlike competitors that blend AI models with traditional numerical weather prediction (NWP), Spire has developed the AI-S2S model entirely in-house, optimizing every component for operational forecasting at scale.

“Spire’s AI-S2S model isn’t just a technological breakthrough — it’s the culmination of world-class AI, meteorology, and computational physics team expertise. With Dr. Acharya leading AI-driven weather and S2S modeling, Dr. Oyedotun advancing machine learning techniques, and Dr. Vela optimizing the computational infrastructure, and help from a diverse team of experts, Spire is setting a new standard for AI-based sub-seasonal forecasting,” said Dr. Tom Gowan, Spire Director of Weather Prediction and AI.

As industries face growing climate uncertainties, Spire’s AI-S2S model empowers businesses with the confidence to navigate the future.

“This is not just an experimental model — it’s an operational product designed to deliver real-world value to industries that rely on accurate long-range forecasts,” said Dr. Acharya. “By seamlessly integrating AI, ensemble forecasting, and satellite data assimilation, Spire is redefining what’s possible in sub-seasonal forecasting.”